Adding the servers to your MCP client

All six servers use the one modelcontextprotocol binary over stdio. Use a separate entry per server. The examples use explicit workspace paths, but the path can be omitted for filesystem, shell, skills, and agents when the client launches MCP server processes in the intended workspace directory.

Shared JSON shape

This conceptual .mcp.json shows all six. ${WORKSPACE} and ${OPENROUTER_API_KEY} are environment-variable references for clients that support such interpolation; otherwise replace the workspace with a literal path and configure the agent credential using that client's supported environment mechanism. No token value belongs in this file.

{
  "mcpServers": {
    "filesystem": { "command": "modelcontextprotocol", "args": ["filesystem", "${WORKSPACE}"] },
    "fetch": { "command": "modelcontextprotocol", "args": ["fetch"] },
    "memory": { "command": "modelcontextprotocol", "args": ["memory"] },
    "shell": { "command": "modelcontextprotocol", "args": ["shell", "${WORKSPACE}"] },
    "skills": { "command": "modelcontextprotocol", "args": ["skills", "${WORKSPACE}"] },
    "agents": {
      "command": "modelcontextprotocol",
      "args": ["agents", "${WORKSPACE}"],
      "env": { "OPENROUTER_API_KEY": "${OPENROUTER_API_KEY}" }
    }
  }
}

skills reads repository instructions without executing them. agents can contact providers and start configured child tools or commands; enable it only in workspaces you trust.

opencode (v1 and v2)

Put this under mcp in global or project opencode.json/opencode.jsonc. opencode local servers use a command array:

{
  "mcp": {
    "filesystem": {
      "type": "local",
      "command": ["modelcontextprotocol", "filesystem", "~/Developer/my-project"],
      "enabled": true,
    },
    "fetch": { "type": "local", "command": ["modelcontextprotocol", "fetch"], "enabled": true },
    "memory": { "type": "local", "command": ["modelcontextprotocol", "memory"], "enabled": true },
    "shell": {
      "type": "local",
      "command": ["modelcontextprotocol", "shell", "~/Developer/my-project"],
      "enabled": true,
    },
    "skills": {
      "type": "local",
      "command": ["modelcontextprotocol", "skills", "~/Developer/my-project"],
      "enabled": true,
    },
    "agents": {
      "type": "local",
      "command": ["modelcontextprotocol", "agents", "~/Developer/my-project"],
      "enabled": true,
    },
  },
}

Claude Desktop

Edit claude_desktop_config.json from Claude's Developer settings (macOS: ~/Library/Application Support/Claude/; Windows: %APPDATA%\Claude\):

{
  "mcpServers": {
    "filesystem": {
      "command": "modelcontextprotocol",
      "args": ["filesystem", "~/Developer/my-project"]
    },
    "fetch": { "command": "modelcontextprotocol", "args": ["fetch"] },
    "memory": { "command": "modelcontextprotocol", "args": ["memory"] },
    "shell": { "command": "modelcontextprotocol", "args": ["shell", "~/Developer/my-project"] },
    "skills": { "command": "modelcontextprotocol", "args": ["skills", "~/Developer/my-project"] },
    "agents": { "command": "modelcontextprotocol", "args": ["agents", "~/Developer/my-project"] }
  }
}

Restart Claude Desktop after editing.

Claude Code

Add each stdio server, or use a project .mcp.json with the shared mcpServers shape:

claude mcp add filesystem -- modelcontextprotocol filesystem ~/Developer/my-project
claude mcp add fetch -- modelcontextprotocol fetch
claude mcp add memory -- modelcontextprotocol memory
claude mcp add shell -- modelcontextprotocol shell ~/Developer/my-project
claude mcp add skills -- modelcontextprotocol skills ~/Developer/my-project
claude mcp add agents -- modelcontextprotocol agents ~/Developer/my-project

For a committed .mcp.json, give each shared entry "type": "stdio".

OpenAI Codex CLI

Add these tables to ~/.codex/config.toml or .codex/config.toml:

[mcp_servers.filesystem]
command = "modelcontextprotocol"
args = ["filesystem", "~/Developer/my-project"]
[mcp_servers.fetch]
command = "modelcontextprotocol"
args = ["fetch"]
[mcp_servers.memory]
command = "modelcontextprotocol"
args = ["memory"]
[mcp_servers.shell]
command = "modelcontextprotocol"
args = ["shell", "~/Developer/my-project"]
[mcp_servers.skills]
command = "modelcontextprotocol"
args = ["skills", "~/Developer/my-project"]
[mcp_servers.agents]
command = "modelcontextprotocol"
args = ["agents", "~/Developer/my-project"]

Pi agent

After pi install npm:pi-mcp-adapter, put the shared mcpServers JSON shape in project .mcp.json or ~/.pi/agent/mcp.json. Pi's shape is the same:

{
  "mcpServers": {
    "filesystem": {
      "command": "modelcontextprotocol",
      "args": ["filesystem", "~/Developer/my-project"]
    },
    "fetch": { "command": "modelcontextprotocol", "args": ["fetch"] },
    "memory": { "command": "modelcontextprotocol", "args": ["memory"] },
    "shell": { "command": "modelcontextprotocol", "args": ["shell", "~/Developer/my-project"] },
    "skills": { "command": "modelcontextprotocol", "args": ["skills", "~/Developer/my-project"] },
    "agents": { "command": "modelcontextprotocol", "args": ["agents", "~/Developer/my-project"] }
  }
}

Gemini CLI

Use the same mcpServers entries in ~/.gemini/settings.json or .gemini/settings.json:

{
  "mcpServers": {
    "filesystem": {
      "command": "modelcontextprotocol",
      "args": ["filesystem", "~/Developer/my-project"]
    },
    "fetch": { "command": "modelcontextprotocol", "args": ["fetch"] },
    "memory": { "command": "modelcontextprotocol", "args": ["memory"] },
    "shell": { "command": "modelcontextprotocol", "args": ["shell", "~/Developer/my-project"] },
    "skills": { "command": "modelcontextprotocol", "args": ["skills", "~/Developer/my-project"] },
    "agents": { "command": "modelcontextprotocol", "args": ["agents", "~/Developer/my-project"] }
  }
}

Cursor

Use .cursor/mcp.json or ~/.cursor/mcp.json. Cursor supports ${env:VAR} in command, args, and env, so this shape can pass an agent credential:

{
  "mcpServers": {
    "filesystem": {
      "type": "stdio",
      "command": "modelcontextprotocol",
      "args": ["filesystem", "${workspaceFolder}"]
    },
    "fetch": { "type": "stdio", "command": "modelcontextprotocol", "args": ["fetch"] },
    "memory": { "type": "stdio", "command": "modelcontextprotocol", "args": ["memory"] },
    "shell": {
      "type": "stdio",
      "command": "modelcontextprotocol",
      "args": ["shell", "${workspaceFolder}"]
    },
    "skills": {
      "type": "stdio",
      "command": "modelcontextprotocol",
      "args": ["skills", "${workspaceFolder}"]
    },
    "agents": {
      "type": "stdio",
      "command": "modelcontextprotocol",
      "args": ["agents", "${workspaceFolder}"],
      "env": { "OPENROUTER_API_KEY": "${env:OPENROUTER_API_KEY}" }
    }
  }
}

Other clients

Windsurf, Zed, VS Code, Continue, Cline, and Roo Code use the same six binary invocations in their documented MCP configuration shape. Add skills as modelcontextprotocol skills <workspace> and agents as modelcontextprotocol agents <workspace> alongside the existing filesystem, fetch, memory, and shell entries; adapt only the wrapper key (context_servers for Zed, servers for VS Code, and the client's equivalent elsewhere). Use environment references only when that client documents that syntax, and never put a real token in configuration.

Restart the client, then confirm all intended server entries and tools appear.